Brain model

There are broadly two branches to the development of Artificial Intelligence. One, the more successful, seeks to utilise natural problem-solving techniques to solve specific human problems. Examples include things like Genetic Algorithms. The second, somewhat less successful branch, seeks to recreate human intelligence in a computer. This branch has had more luck at giving the impression of human intelligence rather than actually creating it (and, indeed, one may ask quite what is meant by 'intelligence'). Between these two branches lies Neural Networks, which utilise models of the human brain to solve very specific jobs, usually centering on classification or prediction (in fact the two are the same, as prediction is just a classification of the future).

For this project, you could:

Day one: build a perceptron, a basic model of a single neuron that makes predictions.

Day two: get the perceptron to read in its data from a file, and write its predictions out.

Day One

Read through the extra materials on neural networks. Have a go at building a 'Pat Butcher Perceptron' from the associated handout (I know -- this was written a while ago!). Start by hardwiring the input test values into a series of arrays.

Day Two

Get the perceptron so it works by reading in files and so it also writes out predictions. You could then adapt the perceptron away from the ridiculous Pat Butcher example, so it does something using numbers *between* zero and one, rather than zero or one. This would be more useful for things like predicting river floods from current river levels and rainfall. Usually any data would be re-ranged between one and zero when it came in, so you could add in your reranging method from the practical.